deep feedforward networks

E25122

artificial neural network architecture deep learning model

Deep feedforward networks are a class of neural network architectures in which information flows in one direction through multiple layers to learn complex input–output mappings without recurrent connections.

Statements (49)

Predicate	Object
instanceOf	artificial neural network architecture → deep learning model →
canUseActivationFunction	ReLU → leaky ReLU → sigmoid → softmax in output layer for classification → tanh →
canUseLossFunction	cross-entropy loss → mean squared error →
canUseOptimizer	Adam → RMSProp → SGD →
differsFrom	convolutional neural networks → recurrent neural networks →
hasAlternativeName	deep MLPs → deep feedforward neural networks → deep multilayer perceptrons →
hasComponent	input layer → one or more hidden layers → output layer →
hasKeyProperty	composed of layers of units with learnable weights → information flows in one direction → learn complex input–output mappings → multiple hidden layers → no recurrent connections →
hasProperty	depth enables hierarchical feature learning → differentiable with respect to parameters → feedforward computation from inputs to outputs → parameters organized in layers → universal function approximator under mild conditions →
introducedInContextOf	deep learning →
isSubclassOf	feedforward neural networks →
mayUse	batch normalization → residual connections →
regularizedBy	dropout → early stopping → weight decay →
requires	labeled training data for supervised tasks →
trainedBy	supervised learning →
trainedWith	backpropagation → gradient descent → mini-batch gradient descent → stochastic gradient descent →
usedFor	classification → function approximation → pattern recognition → regression → representation learning →
uses	nonlinear activation functions →

Referenced by (1)

Subject (surface form when different)	Predicate
Deep Learning (book) →	subject